Setup


In [1]:
# @title connect to drive
# from google.colab import drive
# drive.mount('/content/drive')
In [2]:
# @title download dataset
!unzip mini_dataset.zip
DATA_DIR = "mini_dataset"
Archive: mini_dataset.zip inflating: mini_dataset/STR/STR_960.png inflating: mini_dataset/TUM/TUM_981.png inflating: mini_dataset/TUM/TUM_95.png inflating: mini_dataset/LYM/LYM_876.png inflating: mini_dataset/DEB/DEB_974.png inflating: mini_dataset/LYM/LYM_8.png inflating: mini_dataset/DEB/DEB_99.png inflating: mini_dataset/NOR/NOR_862.png inflating: mini_dataset/ADI/ADI_985.png inflating: mini_dataset/MUS/MUS_967.png inflating: mini_dataset/MUC/MUC_961.png inflating: mini_dataset/ADI/ADI_899.png inflating: mini_dataset/NOR/NOR_734.png inflating: mini_dataset/MUC/MUC_97.png inflating: mini_dataset/STR/STR_972.png inflating: mini_dataset/MUS/MUS_889.png
In [3]:
# @title display images
import os
import matplotlib.pyplot as plt
for class_dir in os.listdir(DATA_DIR):
    class_dir_path = os.path.join(DATA_DIR, class_dir)
    print(class_dir)
    for img in os.listdir(class_dir_path):
        img_path = os.path.join(class_dir_path, img)
        if not os.path.isfile(img_path):
            continue
        plt.imshow(plt.imread(img_path))
        plt.show()
ADI
Output
Output
MUC
Output
Output
NOR
Output
Output
LYM
Output
Output
DEB
Output
Output
MUS
Output
Output
TUM
Output
Output
STR
Output
Output
In [4]:
# @title install libraries + settings
import os
os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
RNG_SEED = 123

!pip install evaluate
!unzip checkpoint-13000.zip
Defaulting to user installation because normal site-packages is not writeable Requirement already satisfied: evaluate in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (0.4.6) Requirement already satisfied: datasets>=2.0.0 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from evaluate) (4.4.1) Requirement already satisfied: numpy>=1.17 in /Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/site-packages (from evaluate) (1.24.2) Requirement already satisfied: dill in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from evaluate) (0.4.0) Requirement already satisfied: pandas in /Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/site-packages (from evaluate) (1.5.3) Requirement already satisfied: requests>=2.19.0 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from evaluate) (2.32.3) Requirement already satisfied: tqdm>=4.62.1 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from evaluate) (4.67.1) Requirement already satisfied: xxhash in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from evaluate) (3.6.0) Requirement already satisfied: multiprocess in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from evaluate) (0.70.18) Requirement already satisfied: fsspec>=2021.05.0 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from fsspec[http]>=2021.05.0->evaluate) (2025.3.0) Requirement already satisfied: huggingface-hub>=0.7.0 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from evaluate) (0.30.2) Requirement already satisfied: packaging in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from evaluate) (24.2) Requirement already satisfied: filelock in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from datasets>=2.0.0->evaluate) (3.12.4) Requirement already satisfied: pyarrow>=21.0.0 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from datasets>=2.0.0->evaluate) (22.0.0) Requirement already satisfied: httpx<1.0.0 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from datasets>=2.0.0->evaluate) (0.27.2) Requirement already satisfied: pyyaml>=5.1 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from datasets>=2.0.0->evaluate) (6.0.2) Requirement already satisfied: aiohttp!=4.0.0a0,!=4.0.0a1 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from fsspec[http]>=2021.05.0->evaluate) (3.11.18) Requirement already satisfied: typing-extensions>=3.7.4.3 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from huggingface-hub>=0.7.0->evaluate) (4.12.2) Requirement already satisfied: charset-normalizer<4,>=2 in /Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/site-packages (from requests>=2.19.0->evaluate) (3.1.0) Requirement already satisfied: idna<4,>=2.5 in /Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/site-packages (from requests>=2.19.0->evaluate) (3.4) Requirement already satisfied: urllib3<3,>=1.21.1 in /Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/site-packages (from requests>=2.19.0->evaluate) (1.26.15) Requirement already satisfied: certifi>=2017.4.17 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from requests>=2.19.0->evaluate) (2024.8.30) Requirement already satisfied: python-dateutil>=2.8.1 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from pandas->evaluate) (2.9.0.post0) Requirement already satisfied: pytz>=2020.1 in /Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/site-packages (from pandas->evaluate) (2022.7.1) Requirement already satisfied: aiohappyeyeballs>=2.3.0 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from aiohttp!=4.0.0a0,!=4.0.0a1->fsspec[http]>=2021.05.0->evaluate) (2.6.1) Requirement already satisfied: aiosignal>=1.1.2 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from aiohttp!=4.0.0a0,!=4.0.0a1->fsspec[http]>=2021.05.0->evaluate) (1.3.2) Requirement already satisfied: async-timeout<6.0,>=4.0 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from aiohttp!=4.0.0a0,!=4.0.0a1->fsspec[http]>=2021.05.0->evaluate) (5.0.1) Requirement already satisfied: attrs>=17.3.0 in /Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/site-packages (from aiohttp!=4.0.0a0,!=4.0.0a1->fsspec[http]>=2021.05.0->evaluate) (22.2.0) Requirement already satisfied: frozenlist>=1.1.1 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from aiohttp!=4.0.0a0,!=4.0.0a1->fsspec[http]>=2021.05.0->evaluate) (1.6.0) Requirement already satisfied: multidict<7.0,>=4.5 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from aiohttp!=4.0.0a0,!=4.0.0a1->fsspec[http]>=2021.05.0->evaluate) (6.4.3) Requirement already satisfied: propcache>=0.2.0 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from aiohttp!=4.0.0a0,!=4.0.0a1->fsspec[http]>=2021.05.0->evaluate) (0.3.1) Requirement already satisfied: yarl<2.0,>=1.17.0 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from aiohttp!=4.0.0a0,!=4.0.0a1->fsspec[http]>=2021.05.0->evaluate) (1.20.0) Requirement already satisfied: anyio in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from httpx<1.0.0->datasets>=2.0.0->evaluate) (4.9.0) Requirement already satisfied: httpcore==1.* in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from httpx<1.0.0->datasets>=2.0.0->evaluate) (1.0.5) Requirement already satisfied: sniffio in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from httpx<1.0.0->datasets>=2.0.0->evaluate) (1.3.1) Requirement already satisfied: h11<0.15,>=0.13 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from httpcore==1.*->httpx<1.0.0->datasets>=2.0.0->evaluate) (0.14.0) Requirement already satisfied: six>=1.5 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from python-dateutil>=2.8.1->pandas->evaluate) (1.17.0) Requirement already satisfied: exceptiongroup>=1.0.2 in /Users/vidhya/Library/Python/3.10/lib/python/site-packages (from anyio->httpx<1.0.0->datasets>=2.0.0->evaluate) (1.2.2) [notice] A new release of pip is available: 24.3.1 -> 25.3 [notice] To update, run: pip install --upgrade pip Archive: checkpoint-13000.zip inflating: checkpoint-13000/training_args.bin inflating: checkpoint-13000/scheduler.pt inflating: checkpoint-13000/trainer_state.json inflating: checkpoint-13000/rng_state.pth inflating: checkpoint-13000/config.json inflating: checkpoint-13000/model.safetensors inflating: checkpoint-13000/optimizer.pt
In [5]:
# @title load data
import glob
import pandas as pd
from datasets import load_dataset, DatasetDict


dataset = load_dataset("imagefolder", data_dir=DATA_DIR)
/Users/vidhya/Library/Python/3.11/lib/python/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html from .autonotebook import tqdm as notebook_tqdm
In [6]:
# @title Preprocess Data

from torchvision.transforms import Compose, ToTensor, Resize, Normalize, CenterCrop

_transform_pipeline = Compose([
    Resize(256),
    ToTensor(),
])

def transform_img(batch):
    batch["pixel_values"] = [ _transform_pipeline(img.convert("RGB")) for img in batch["image"]]
    return batch

dataset = dataset.map(
    transform_img,batched=True,batch_size=32,remove_columns=["image"]
)
In [7]:
# @title Collate Function

from transformers import DefaultDataCollator

collator = DefaultDataCollator()
In [10]:
# @title Testing
import torch
from transformers import TrainingArguments, Trainer, EarlyStoppingCallback, AutoModelForImageClassification

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

import numpy as np
import evaluate
import numpy as np
import torch

roc_auc_metric = evaluate.load("roc_auc", "multiclass")
accuracy_metric = evaluate.load("accuracy")
precision_metric = evaluate.load("precision")
f1_metric =evaluate.load("f1")
recall_metric =evaluate.load("recall")

def compute_metrics(eval_pred):
    logits, labels = eval_pred

    probs = torch.softmax(torch.tensor(logits, dtype=torch.float32), dim=1).numpy()
    labels = labels.astype(np.int32)
    predictions = np.argmax(logits, axis=-1)

    results = {}

    results["roc_auc"] = roc_auc_metric.compute(
        references=labels,
        prediction_scores=probs,
        multi_class="ovr",
    )["roc_auc"]

    results["accuracy"] = accuracy_metric.compute(
        predictions=predictions,
        references=labels,
    )["accuracy"]

    results["precision"] = precision_metric.compute(
        predictions=predictions,
        references=labels,
        average="macro",
    )["precision"]

    results["f1"] = f1_metric.compute(
        predictions=predictions,
        references=labels,
        average="macro",
    )["f1"]

    results["recall"] = recall_metric.compute(
        predictions=predictions,
        references=labels,
        average="macro",
    )["recall"]

    return results


checkpoint_path = "./checkpoint-13000"

model = AutoModelForImageClassification.from_pretrained(
    checkpoint_path,
    use_safetensors=True
)

training_args = TrainingArguments(
    eval_strategy = "steps",
    save_strategy = "steps",
    learning_rate = 1e-4,
    weight_decay = 1e-4,
    num_train_epochs = 10,
    save_steps=500,
    eval_steps=500,
    load_best_model_at_end=True,
    metric_for_best_model="eval_loss",
    greater_is_better=False,
    logging_steps = 500,
    report_to = "tensorboard",
    save_total_limit=3,
    remove_unused_columns=False
    )

trainer = Trainer(
    model=model,
    data_collator = collator,
    train_dataset=dataset,
    args=training_args,
    eval_dataset=dataset,
    compute_metrics=compute_metrics,
    callbacks=[EarlyStoppingCallback(early_stopping_patience=5)],
)


test_results = trainer.evaluate(dataset)
print(test_results)

pd.DataFrame(list(test_results.items()), columns=["Metric", "Value"])
/Users/vidhya/Library/Python/3.11/lib/python/site-packages/torch/utils/data/dataloader.py:692: UserWarning: 'pin_memory' argument is set as true but not supported on MPS now, device pinned memory won't be used. warnings.warn(warn_msg)
[2/2 00:00]
/Users/vidhya/Library/Python/3.11/lib/python/site-packages/sklearn/metrics/_classification.py:1731: UndefinedMetricWarning: Precision is ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior. _warn_prf(average, modifier, f"{metric.capitalize()} is", result.shape[0]) early stopping required metric_for_best_model, but did not find eval_loss so early stopping is disabled
{'eval_train_loss': 0.8428382873535156, 'eval_train_model_preparation_time': 0.0024, 'eval_train_roc_auc': 0.9598214285714286, 'eval_train_accuracy': 0.625, 'eval_train_precision': 0.6541666666666666, 'eval_train_f1': 0.5839285714285714, 'eval_train_recall': 0.625, 'eval_train_runtime': 6.4443, 'eval_train_samples_per_second': 2.483, 'eval_train_steps_per_second': 0.31}
Metric Value
0 eval_train_loss 0.842838
1 eval_train_model_preparation_time 0.002400
2 eval_train_roc_auc 0.959821
3 eval_train_accuracy 0.625000
4 eval_train_precision 0.654167
5 eval_train_f1 0.583929
6 eval_train_recall 0.625000
7 eval_train_runtime 6.444300
8 eval_train_samples_per_second 2.483000
9 eval_train_steps_per_second 0.310000